[pylrn]

Summary

Adds complete typed support for Speech Understanding enhancements across all feature requests:

• effort parameter: Supported across SpeakerIdentificationRequest, TranslationRequest, CustomFormattingRequest, and harmonized with upstream SummarizationRequest and SummarizationResponse.
• force_translation parameter: Added to TranslationRequest to enable translating audio even when the detected source language matches a target language.
• format_utterances parameter: Added to CustomFormattingRequest to format utterance-level text.
• speakers / SpeakerDescription: Added SpeakerDescription model with dictionary compatibility (__getitem__, get, __contains__) and typed speakers on SpeakerIdentificationRequest.
• SpeechUnderstandingEffort: Enum (low, medium) exported in assemblyai and assemblyai.types and included in __all__.
• Rebase onto upstream master: Fully rebased on upstream master (including feat(pysdk): Typed support for SU understanding transcript config api #274) with version bumped to 1.6.6 for CI release checks.

Restores full feature and typing parity with assemblyai-node-sdk (PR #197) and assemblyai-mcp (PR #5).

Changes

• assemblyai/types.py:
  • Define SpeechUnderstandingEffort enum (low, medium).
  • Define SpeakerDescription model supporting both attribute and dict-item subscripting with open extra properties.
  • Expose speakers: Optional[List[Union[SpeakerDescription, Dict[str, Any]]]] = None and effort on SpeakerIdentificationRequest.
  • Expose force_translation: Optional[bool] = False and effort on TranslationRequest.
  • Expose format_utterances: Optional[bool] = False and effort on CustomFormattingRequest.
  • Type effort on SummarizationRequest and SummarizationResponse to accept SpeechUnderstandingEffort enum or literal string.
• assemblyai/__init__.py: Export SpeechUnderstandingEffort and SpeakerDescription in __all__.
• assemblyai/__version__.py: Bump to 1.6.6.
• CLAUDE.md: Document speech_understanding under TranscriptionConfig options.
• tests/unit/test_speech_understanding.py: 13 comprehensive unit tests covering enum values, request models, dictionary serialization with exclude_none=True, TranscriptionConfig assignment, SpeakerDescription, format_utterances, and response parsing.

Testing

• Full unit test suite passes: 754 passed in 38s.
• ruff check and ruff format --check pass cleanly on all modified files.